On Wed, Jun 28, 2017 at 04:34:18PM -0500, Josh Poimboeuf wrote: > On Wed, Jun 28, 2017 at 04:25:19PM -0500, Josh Poimboeuf wrote: > > On Wed, Jun 28, 2017 at 02:02:09PM -0400, Jason Baron wrote: > > > Hi, > > > > > > If I have kernel source S, apply a kpatch module A, but then want to > > > 'revert' parts of A, by applying a cumulative patch B, such that some > > > functions are no longer different from source kernel S, how do I do that? > > > > > > It seems like there are a few options: > > > > > > 1) remove or disable module A, then load module B. This leaves an > > > un-patched window and rmmod isn't possible without reliable backtrace. > > > I'm also not sure this works with the kpatch 'stacking'. > > > > > > 2) modify patch B such that any functions that are initially unchanged > > > from S, are in fact modified but modified in an acceptable way. > > > > > > 3) maybe the kernel can understand that are not added to or modified by > > > a cumulative patch should be reverted back to their original state? > > > > > > #3 seems to me to be the cleanest...but maybe I'm missing something... > > > > Hi Jason, > > > > We don't have a mechanism for reverting patches, so #2 will probably be > > your best bet. > > Sorry, that was both poorly worded and incorrect, let me try again :-) > > I meant to say we don't have a mechanism for atomically replacing the > entire contents of patch A with patch B. > > But I was mistaken. Technically we do have a "replace" mechanism. It > was deprecated, see https://github.com/dynup/kpatch/issues/456 for more > details. But it should work in theory if you load the patch module > manually with insmod and the 'replace=1' argument. We had some issues > with it though, so #2 might be a safer bet.
And I should note that the replace only works with kpatch.ko, not with livepatch. We may add replace support to livepatch eventually. -- Josh _______________________________________________ kpatch mailing list [email protected] https://www.redhat.com/mailman/listinfo/kpatch
